A Novel LLM-Driven Pipeline for Digital Twin Generation Using Clinical Notes to Inform Medical Recommendations

Overview

Clinical decision-making relies heavily on patient documentation, rich in temporal detail, context, and clinician reasoning, contained in electronic health records, yet much of this information remains underutilized due to its unstructured format. This study presents a digital twins pipeline that leverages large language models (LLMs) and embedding-based similarity search to transform raw clinical notes into accurate and structured patient representations to inform data-driven recommendations. Our central hypothesis is that digital twins identified from standardized summaries, extracted clinical timelines, and vector embeddings of prior patient trajectories can generate clinical recommendations that approximate those of trained medical professionals. The pipeline first uses LLMs to clean and standardize historical notes before extracting structured longitudinal timelines capturing events and outcomes. These extractions are embedded using a text-embedding model to produce vectorized representations of patients' clinical trajectories. For a new patient, notes are processed identically and compared to historical embeddings via cosine similarity to identify the most clinically analogous cases, whose trajectories and outcomes inform recommendations. Validation is conducted using the publicly available large-scale MIMIC-IV dataset and corroboration with trained medical professionals. Ultimately, the developed pipeline efficiently delivers clinical recommendations largely in alignment with clinical decisions made by physicians and trained medical professionals, revealing the potential for LLM-driven systems to assist in streamlining clinical decisions and, consequently, to reduce cost and improve quality of care.

Competition history

• CSEF 2026 Medicine & Physiology (Track 2) (Senior Division) · Entry S-21-27

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Biomedical and Health Sciences projects
• CSEF Biomedical and Health Sciences projects
• Every CSEF 2026 project